CN115803783A

CN115803783A - Reconstruction of 3D object models from 2D images

Info

Publication number: CN115803783A
Application number: CN202180044497.5A
Authority: CN
Inventors: 乔治斯·帕潘德里欧; 亚松·柯克基诺斯
Original assignee: Snap Inc
Current assignee: Snap Inc
Priority date: 2020-06-22
Filing date: 2021-06-22
Publication date: 2023-03-14
Also published as: EP4168997A1; GB2598452A; WO2021262667A1; GB202108595D0; GB2598452B; US20230267687A1; KR20230027237A

Abstract

Systems and methods for reconstructing a 3D model of a human body from 2D images are provided that take into account perspective and/or distortion effects. The system and method include reconstructing a three-dimensional model of an object in a three-dimensional scene from a two-dimensional image, the two-dimensional image including an image of the object. The system and method include determining absolute depths of keypoints of an object in the image; determining a three-dimensional position of the keypoint in the three-dimensional scene using the absolute depth of the keypoint; generating a three-dimensional representation of the object using a neural network, the three-dimensional representation including mesh nodes defined in a coordinate system relative to the keypoints; and locating a three-dimensional representation of the object in the scene based on the locations of the keypoints by applying a position-dependent rotation to the three-dimensional object.

Description

Reconstruction of 3D object models from 2D images

PRIORITY CLAIM

This application claims priority from uk application serial No. 2009515.4, filed on 22/6/2020 and U.S. patent application serial No. 17/249,441, filed on 2/3/2021, each of which is incorporated herein by reference in its entirety.

Technical Field

This specification relates to reconstructing a 3D model of a human from 2D images. In particular, the present description relates to systems and methods for reconstructing a 3D model of a human body from 2D images, which take into account perspective and/or distortion effects.

Background

Reconstructing three-dimensional (3D) models of humans from two-dimensional (2D) images is a common problem in computer vision.

Brief description of several views of the drawings

In the drawings, which are not necessarily drawn to scale, like reference numerals may describe similar parts throughout the different views. To readily identify the discussion of any particular element or act, one or more of the most significant digits of a reference number refer to the figure number in which that element is first introduced. Some non-limiting examples are shown in the figures of the accompanying drawings, in which:

fig. 1 is a diagrammatic representation of a networked environment in which the present disclosure may be deployed, according to some examples.

Fig. 2 is a diagrammatic representation of a messaging system having both client-side functionality and server-side functionality, according to some examples.

Fig. 3 is a diagrammatic representation of a data structure as maintained in a database, according to some examples.

Fig. 4 is a diagrammatic representation of a message according to some examples.

Fig. 5 illustrates an overview of the geometry of a camera device capturing two-dimensional images of objects in a three-dimensional scene, according to some examples.

Fig. 6A illustrates an example of a method of reconstructing a three-dimensional object in a three-dimensional scene from a two-dimensional image, according to some examples.

Fig. 6B illustrates an example of a method for training a neural network to correctly reconstruct a three-dimensional representation of an object from two-dimensional images, according to some examples.

Fig. 7 illustrates an example of a method for estimating a three-dimensional pose of an object in an image, according to some examples.

Fig. 8 illustrates an example of another method for estimating a three-dimensional pose of an object in an image, according to some examples.

Fig. 9 illustrates an example of a variation of the method in fig. 8, according to some examples.

Fig. 10 is a flow diagram illustrating example operations of a messaging application server according to an example.

Fig. 11 is a diagrammatic representation of machine in the form of a computer system within which a set of instructions, for causing the machine to perform any one or more of the methodologies discussed herein, may be executed, according to some examples.

FIG. 12 is a block diagram illustrating a software architecture in which examples may be implemented.

Detailed Description

The following description includes systems, methods, techniques, instruction sequences, and computer program products that embody illustrative examples of the present disclosure. In the following description, for purposes of explanation, numerous specific details are set forth in order to provide an understanding of the various examples. It will be apparent, however, to one skilled in the art that the examples may be practiced without these specific details. In general, well-known instruction instances, protocols, structures, and techniques need not be shown in detail.

In the current literature, the method of reconstructing a 3D human model is validated by comparing a model-based reconstruction within an image window with a 2D image using scaled forward projections. However, this is a rough approximation, which is only valid if all the body of the person is located at the same depth from the camera, e.g. like a cardboard parallel to the image plane. As soon as a person comes closer to the camera (which is common for "self-portrait" photographs), the perspective distortion itself becomes apparent. In these cases, the hand, head, or nose holding the imaging device is visually enlarged relative to the farther torso. This may result in a failure of the orthogonal approximation, resulting in 3D head/hand reconstruction being significantly smaller than their image counterparts. In addition to this particular use case, for downstream augmented reality applications, it is necessary to incorporate perspective projections, where the person needs to be correctly positioned within the environment reconstructed using the metric coordinates.

In general, this specification relates to systems and methods for reconstructing a 3D model of a human body from 2D images, which are capable of dealing with perspective and/or distortion effects. The systems and methods described herein may allow for more accurate reconstruction of 3D objects from 2D images than current methods. The present description relates to a system and method for 3D interpretation of humans in the case of RGB images. Several aspects may be used to perform this task; these aspects may be used alone or in combination. In one aspect, absolute depths of keypoints of an object in an image are determined, and the absolute depths are used to determine three-dimensional locations of the keypoints. A 3D representation of the object is generated using a neural network, and the 3D representation of the object is positioned in the 3D scene based on the 3D locations of the keypoints. In one implementation, a position-dependent rotation is applied to a 3D object to position a 3D representation of the object in a scene.

The location of the keypoints in the 3D scene may be determined by: an intersection of a ray connecting a camera capturing a two-dimensional image to a keypoint of the object and an image plane of the two-dimensional image is determined using the same neural network or another neural network, and a position of the keypoint of the object in a coordinate system of the three-dimensional scene is determined based on the world depth and the intersection. The location of the keypoints in the three-dimensional scene may be further determined based on the focal distance of the camera device capturing the two-dimensional image.

In one implementation, the located three-dimensional representation may be projected into a two-dimensional image plane to generate a projection image of the object. The parameters of the neural network may be updated based on comparing the location of a point in the projection image with the location of a corresponding point in the two-dimensional image.

In another aspect, a three-dimensional model of a human (e.g., an object) may be reconstructed in a three-dimensional scene from a two-dimensional image of the human. The three-dimensional model may comprise a directed graph, wherein for each node in the directed graph: a 3D offset field associated with an edge connecting a node to a previous node in the directed graph is estimated using one or more convolutional neural networks and based on two-dimensional image coordinates of the previous node. Disparity is estimated in the vicinity of a point associated with a previous node. The three-dimensional coordinates of the previous node, the three-dimensional offset field, and the estimated disparity are used to determine the three-dimensional coordinates of the node.

In some implementations, the three-dimensional coordinates of the nodes may be estimated by: the two-dimensional image coordinates of the node are determined based on the two-dimensional image coordinates of the previous node and the two components of the three-dimensional offset field. The depth coordinate of the node may be determined based on the estimated disparity and one of the depth coordinate of the previous node, the component of the three-dimensional offset field. The two-dimensional image coordinates of the nodes may be converted to two-dimensional scene coordinates of the nodes using the depth coordinates of the nodes or the estimated disparity associated with the nodes.

In some implementations, the depth coordinate of a node may be determined based on the depth coordinate of a previous node by: an initial estimate of the depth coordinate is determined based on the depth coordinate of the previous node, one of the components of the three-dimensional offset field, and the estimated disparity. The initial estimate of the depth coordinate may be refined by translating the nodes in the depth direction by a constant value representing a weighted average of differences between the depth coordinate of each node and the disparity estimate of the depth of each node.

In some cases, the one or more convolutional neural networks may include: a first convolutional neural network configured to estimate a two-dimensional offset field in a plane associated with the two-dimensional image; and a second neural network configured to estimate a one-dimensional depth offset field in a direction perpendicular to the plane.

In another aspect, a three-dimensional model of a human (e.g., an object) may be reconstructed from the two-dimensional images. The three-dimensional model may include a joint having a joint position and a joint rotation. Known joint positions for each joint are received and, starting from the root joint, the joint pose for that joint is determined based on a recurrence relation that defines the joint pose for that joint as a function of the joint pose for the previous joint, a predefined offset between that joint and the previous joint, a bone rotation, and a bone scaling. Bone rotation (and in some examples, scaling/translation) is determined based at least in part on the known joint position of each joint. The known joint locations may include, for example, actual real joint locations determined using a motion capture process. The known joint positions may be estimated joint positions obtained using a convolutional neural network (e.g., a network of 3D positions trained to identify joint positions from 2D images).

The known joint positions may be estimated by combining the joint positions estimated by the one or more object part models with the estimated three-dimensional keypoint positions of the object. Keypoint locations may be estimated, for example, by CNN or disparity-based keypoint localization methods. The object part model may be a model that is dedicated to determining a 3D model of a particular object part (e.g., a hand model, a head model, a torso model, etc.). The subject part model may be a CNN trained to generate a 3D representation (e.g., a mesh) of the part of the subject from a 2D image of the subject.

At least one of the joints in the three-dimensional model may be a joint having a single sub-joint, and the bone scaling of each such joint is determined based on a predefined length between the joint and the sub-joint of the joint. The known length between the joint and the sub-joints of the joint may be determined based on a difference between the known position of the joint and the known positions of the sub-nodes of the joint.

In one example, joint rotation R for joint j _j The recursive relationship may be given by:

R _j ＝R _i E _j s _j B _j

wherein R is _i Is the joint rotation of the previous joint i, E _j Is a predetermined rotational offset, s, between joint i and joint j _j Is the bone scaling of joint j, and B _j Is the bone rotation of joint j; and, joint position t for joint j _j ，

t _j ＝t _i +R _i d _j

Wherein, t _i Is the joint position of the previous joint i, and d _j Is the predetermined position offset between joint i and joint j.

According to another aspect of the present specification, an apparatus is described, the apparatus comprising one or more processors and a memory storing computer readable code which, when executed by the one or more processors, causes the system to perform a method according to any of the methods described herein.

According to another aspect of the present specification, a computer program product comprising computer code which, when executed by a computing device, causes the computing device to perform any of the methods described herein is described.

Networked computing environment

Fig. 1 is a block diagram illustrating an example messaging system 100 for exchanging data (e.g., messages and associated content) over a network. The messaging system 100 includes multiple instances of the client device 102, each of which hosts multiple applications including a messaging client 104 and other external applications 109 (e.g., third party applications). Each messaging client 104 is communicatively coupled to the messaging client 104 (e.g., hosted on a respective other client device 102), the messaging server system 108, and other instances of the external application server 110 via a network 112 (e.g., the internet). The messaging client 104 can also communicate with the locally hosted third party application 109 using an Application Program Interface (API).

The messaging client 104 is capable of communicating and exchanging data with other messaging clients 104 and the messaging server system 108 via the network 112. Data exchanged between messaging clients 104 and messaging server system 108 includes functions (e.g., commands to activate the functions) and payload data (e.g., text, audio, video, or other multimedia data).

The messaging server system 108 provides server-side functionality to the particular messaging client 104 via the network 112. While certain functions of messaging system 100 are described herein as being performed by messaging client 104 or by messaging server system 108, the positioning of certain functions within messaging client 104 or messaging server system 108 may be a design choice. For example, it may be technically preferable: certain techniques and functionality are initially deployed within messaging server system 108, but later migrated to messaging client 104 if client device 102 has sufficient processing power.

Messaging server system 108 supports various services and operations provided to messaging client 104. Such operations include sending data to the messaging client 104, receiving data from the messaging client 104, and processing data generated by the messaging client 104. The data may include message content, client device information, geo-location information, media enhancements and overlays, message content persistency conditions, social networking information, and live event information, as examples. Data exchange within messaging system 100 is motivated and controlled through functionality available via a User Interface (UI) of messaging client 104.

Turning now specifically to messaging server system 108, an Application Program Interface (API) server 116 is coupled to application server 114 and provides a programming interface to application server 114. Application server 114 is communicatively coupled to database server 120, and database server 120 facilitates access to a database 126, the database 126 storing data associated with messages processed by application server 114. Similarly, a web server 128 is coupled to the application server 114 and provides a web-based interface to the application server 114. To this end, the web server 128 processes incoming network requests via the hypertext transfer protocol (HTTP) and several other related protocols.

An Application Program Interface (API) server 116 receives and sends message data (e.g., commands and message payloads) between the client device 102 and the application server 114. In particular, an Application Program Interface (API) server 116 provides a set of interfaces (e.g., routines and protocols) that messaging client 104 may call or query to invoke the functionality of application server 114. An Application Program Interface (API) server 116 exposes various functions supported by the application server 114, including: registering an account; a login function; sending a message from a particular messaging client 104 to another messaging client 104 via an application server 114; sending a media file (e.g., an image or video) from messaging client 104 to messaging server 118; and to enable access by another messaging client 104, set a collection of media data (e.g., stories), retrieve a friends list of a user of the client device 102, retrieve such a collection, retrieve messages and content, add and delete entities (e.g., friends) to and from an entity graph (e.g., a social graph), locations of friends in the social graph, and open application events (e.g., related to the messaging client 104).

The application server 114 hosts a number of server applications and subsystems, including, for example, a messaging server 118, an image processing server 122, and a social network server 124. Messaging server 118 implements a number of message processing techniques and functions that relate, inter alia, to the aggregation and other processing of content (e.g., text and multimedia content) included in messages received from multiple instances of messaging client 104. As will be described in further detail, text and media content from multiple sources may be aggregated into a collection of content (e.g., referred to as a story or gallery). These collections are then made available to messaging client 104. Such processing may also be performed on the server side by the messaging server 118, taking into account the hardware requirements for other processor and memory intensive processing of the data.

Application server 114 also includes an image processing server 122, which image processing server 122 is dedicated to performing various image processing operations, typically with respect to images or video within the payload of messages sent from messaging server 118 or received at messaging server 118. The detailed functions of the image processing server 122 are shown and described in connection with fig. 5 through 10.

In one example, the image processing server 122 implements one or more neural networks (e.g., convolutional neural networks-CNN) trained to regress the absolute depth and/or disparity of objects (such as humans) in an image, i.e., the amount that the focal length of the camera that captured the image is invariant. This allows the CNN to be trained with images from cameras of different focal lengths. Under test, the camera focal length (given or estimated) may be used to convert the regressed absolute depth/disparity back to the measured depth.

In some examples, methods are introduced to promote rigid objects to 3D beyond weak perspective. For example, the image processing server 122 may introduce the following corrections: this correction mitigates the distortion effect that arises as one gets farther from the center of the camera. Alternatively or additionally, the vertex disparity estimation may be used to accurately re-project to the 2D coordinates. In some examples, a self-calibration method for focus estimation is used, where minimization of the 2D re-projection objective delivers an estimate of the focus of the image capture device. This allows matching the parametric model to the image with significantly lower error, delivering both a better fit and a better surveillance signal for distillation-based surveillance. In some examples, a site-based 3D skeleton estimation approach is used that handles depth variability within the same object by using locally constant but globally varying disparity estimation. This may allow for severe foreshortening effects, allowing for robust processing in case of objects (e.g. hands or faces) close to the camera.

In some examples, closed inverse kinematics solutions may be used to narrow the gap between parametric model-based (i.e., "top-down") methods of monocular 3D body reconstruction and CNN-based (or other "bottom-up" type models) predictions of 3D body joints. The method may incorporate hands, allowing for obtaining a model-based interpretation of body-hand pixels that is accurate. These examples may be used for the results of fitting a parametric model of the body subject to physical constraints on joint position and orientation to one or more bottom-up models that generally do not contain such constraints. Furthermore, by fitting a parametric model of the entire body to a 3D model of the body part from the bottom-up model, they can be effectively "stitched together".

The social network server 124 supports and makes available various social networking functions and services to the messaging server 118. To this end, the social network server 124 maintains and accesses an entity graph 308 (shown in FIG. 3) within the database 126. Examples of functions and services supported by the social network server 124 include identifying other users in the messaging system 100 that have a relationship with a particular user or that the particular user is "paying attention" to, and identifying interests and other entities of the particular user.

Returning to the messaging client 104, features and functionality of external resources (e.g., third party applications 109 or applets) are available to the user via an interface of the messaging client 104. The messaging client 104 receives a user selection of an option to launch or access a feature of an external resource (e.g., a third party resource) (e.g., an external application 109). The external resource may be a third party application (external application 109) (e.g., a "native application") installed on the client device 102, or a small scale version of the third party application (e.g., a "applet") hosted on the client device 102 or remote from the client device 102 (e.g., on the third party server 110). The small-scale version of the third-party application includes a subset of the features and functionality of the third-party application (e.g., a full-scale local version of the third-party standalone application), and is implemented using a markup language document. In one example, the small-scale version of the third-party application (e.g., the "applet") is a web-based markup language version of the third-party application and is embedded in the messaging client 104. In addition to using markup language documents (e.g., jjs files), applets may also contain scripting languages (e.g., jjs files or json files) and style sheets (e.g., jss files).

In response to receiving a user selection of an option to launch or access a feature of an external resource (external application 109), messaging client 104 determines whether the selected external resource is a web-based external resource or a locally installed external application. In some cases, the external application 109 installed locally on the client device 102 may be launched independently of the messaging client 104 and separately from the messaging client 104, for example, by selecting an icon corresponding to the external application 109 on a home screen of the client device 102. A small-scale version of such an external application may be launched or accessed via messaging client 104, and in some examples, no or limited portions of the small-scale external application may be accessed from outside messaging client 104. The small scale external application may be launched by the messaging client 104 receiving a markup language document associated with the small scale external application from the external application server 110 and processing such document.

In response to determining that the external resource is a locally installed external application 109, the messaging client 104 instructs the client device 102 to launch the external application 109 by executing locally stored code corresponding to the external application 109. In response to determining that the external resource is a web-based resource, the messaging client 104 communicates with the external application server 110 to obtain a markup language document corresponding to the selected resource. The messaging client 104 then processes the obtained markup language document to present the web-based external resource within a user interface of the messaging client 104.

Messaging client 104 may notify a user of client device 102 or other users related to such user (e.g., "friends") of activities occurring in one or more external resources. For example, messaging client 104 may provide notifications to participants of a conversation (e.g., a chat session) in messaging client 104 regarding current or recent use of external resources by one or more members of a user group. One or more users may be invited to join an active external resource or to start a recently used external resource that is not currently active (in the group of friends). The external resource may provide participants in the conversation (each using a respective messaging client 104) with the ability to share items, conditions, states, or locations in the external resource with one or more members of a group of users in the chat session. The shared item may be an interactive chat card with which members of the chat may interact, for example, to launch a corresponding external resource, to view specific information within the external resource, or to bring members of the chat to a particular location or state within the external resource. Within a given external resource, a response message may be sent to the user on messaging client 104. The external resource may selectively include different media items in the response based on the current context of the external resource.

The messaging client 104 can present a list of available external resources (e.g., third party or external applications 109 or applets) to the user to launch or access a given external resource. The list may be presented in a context sensitive menu. For example, icons representing different ones of the external applications 109 (or applets) may change based on how the user launches the menu (e.g., from a conversational interface or from a non-conversational interface).

System architecture

Fig. 2 is a block diagram illustrating further details regarding messaging system 100, according to some examples. In particular, the messaging system 100 is shown to include a messaging client 104 and an application server 114. The messaging system 100 includes a plurality of subsystems supported on the client side by the messaging client 104 and on the server side by the application server 114. These subsystems include, for example, an ephemeral timer system 202, an aggregation management system 204, an augmentation system 208, a mapping system 210, a gaming system 212, and an external resource system 220.

The ephemeral timer system 202 is responsible for enforcing temporary or time-limited access to content by the messaging client 104 and the messaging server 118. The ephemeral timer system 202 includes a plurality of timers that selectively enable access (e.g., for rendering and displaying) of messages and associated content via the messaging client 104 based on a duration and display parameters associated with the message or set of messages (e.g., a story). Additional details regarding the operation of the ephemeral timer system 202 are provided below.

The collection management system 204 is responsible for managing groups or collections of media (e.g., collections of text, image video, and audio data). A collection of content (e.g., messages including images, video, text, and audio) may be organized into an "event gallery" or "event story". Such a set may be made available for a specified period of time, for example, for the duration of an event related to the content. For example, content related to a concert may be made available as a "story" for the duration of the concert. The collection management system 204 may also be responsible for publishing icons to the user interface of the messaging client 104 that provide notification of the existence of a particular collection.

In addition, the collection management system 204 also includes a curation interface 206 that allows the collection manager to manage and curate a particular collection of content 206. For example, curation interface 206 enables event organizers to curate content collections related to a particular event (e.g., delete inappropriate content or redundant messages). In addition, the collection management system 204 employs machine vision (or image recognition techniques) and content rules to automatically curate content collections. In some examples, the user may be paid compensation for including the user-generated content into the collection. In such a case, the collection management system 204 operates to automatically pay such users to use their content.

The enhancement system 208 provides various functionality that enables a user to enhance (e.g., annotate or otherwise modify or edit) media content associated with a message. For example, the enhancement system 208 provides functionality related to generating and publishing media overlays for messages processed by the messaging system 100. The enhancement system 208 is operable to supply media overlays or enhancements (e.g., image filters) to the messaging client 104 based on the geographic location of the client device 102. In another example, the enhancement system 208 is operable to provision the messaging client 104 with a media overlay based on other information, such as social networking information of a user of the client device 102. The media overlay may include audio and visual content and visual effects. Examples of audio and visual content include pictures, text, logos, animations, and sound effects. Examples of visual effects include color overlay. The audio and visual content or visual effects may be applied to media content items (e.g., photos) at the client device 102. For example, the media overlay may include text, graphical elements, or images that may be overlaid on a photograph taken by the client device 102. In another example, the media overlay includes a location identification overlay (e.g., venice beach), a live event name, or a business name overlay (e.g., beach cafe). In another example, the augmentation system 208 uses the geographic location of the client device 102 to identify a media overlay that includes the name of the merchant at the geographic location of the client device 102. The media overlay may include other indicia associated with the merchant. The media overlay may be stored in the database 126 and accessed through the database server 120.

In some examples, the augmentation system 208 provides a user-based publication platform that enables a user to select a geolocation on a map and upload content associated with the selected geolocation. The user may also specify the environment in which particular media overlays should be provided to other users. The augmentation system 208 generates a media overlay including the uploaded content and associates the uploaded content with the selected geographic location.

In other examples, the augmentation system 208 provides a merchant-based publication platform that enables a merchant to select a particular media overlay associated with geolocation via a bidding process. For example, the augmentation system 208 associates the media coverage of the highest bidding merchant with the corresponding geographic location for a predefined amount of time. The augmentation system 208 communicates with the image processing server 122 to automatically select and activate an augmented reality experience related to an image captured by the client device 102. Once the user selects an augmented reality experience while scanning an image using the camera in the user environment, one or more images, videos, or augmented reality graphical elements are retrieved and presented as overlays over the scanned image. In some cases, the camera is switched to a front-facing view (e.g., the front camera of client device 102 is activated in response to activation of a particular augmented reality experience) and images from the front camera of client device 102 instead of the rear camera of client device 102 begin to be displayed at client device 102. One or more images, videos, or augmented reality graphical elements are retrieved and presented as overlays over images captured and displayed by the front facing camera of the client device 102.

The map system 210 provides various geo-location functionality and supports the rendering of map-based media content and messages by the messaging client 104. For example, the map system 210 enables a user icon or avatar (e.g., stored in the profile data 316) to be displayed on a map to indicate the user's current or past locations of "friends" in the context of the map, as well as media content (e.g., a collection of messages including photos and videos) generated by such friends. For example, a message published by a user to the messaging system 100 from a particular geographic location may be displayed to a particular user's "friend" at that particular location in the context of a map on a map interface of the messaging client 104. The user may also share his or her location and status information with other users of the messaging system 100 (e.g., using an appropriate status avatar) via the messaging client 104, where the location and status information is similarly displayed to the selected user in the context of the map interface of the messaging client 104.

The gaming system 212 provides various gaming functions in the context of the messaging client 104. The messaging client 104 provides a game interface that provides a list of available games (e.g., web-based games or web-based applications) that can be initiated by a user in the context of the messaging client 104 and played with other users of the messaging system 100. The messaging system 100 also enables a particular user to invite other users to participate in playing a particular game by issuing invitations to such other users from the messaging client 104. The messaging client 104 also supports both voice messaging and text messaging (e.g., chat) in the context of game play, provides a leaderboard for games, and also supports in-game rewards (e.g., money and items).

External resource system 220 provides an interface for messaging client 104 to communicate with external application server 110 to initiate or access external resources. Each external resource (application) server 110 hosts a small-scale version of, for example, a markup language (e.g., HTML 5) based application or an external application (e.g., a game, utility, payment or ride share application external to the messaging client 104). The messaging client 104 can launch a web-based resource (e.g., an application) by accessing the HTML5 file from an external resource (application) server 110 associated with the web-based resource. In some examples, the application hosted by the external resource server 110 is programmed in JavaScript using a Software Development Kit (SDK) provided by the messaging server 118. The SDK includes an Application Programming Interface (API) having functionality that can be called or activated by the web-based application. In some examples, messaging server 118 includes a JavaScript library that provides given third party resource access rights for certain user data of messaging client 104. HTML5 is used as an example technique for programming games, but applications and resources programmed based on other techniques may be used.

To integrate the functionality of the SDK into the web-based resource, the SDK is downloaded from the messaging server 118 by the external resource (application) server 110 or otherwise received by the external resource (application) server 110. Once downloaded or received, the SDK is included as part of the application code for the web-based external resource. The code of the web-based resource may then invoke or activate certain functions of the SDK to integrate features of the messaging client 104 into the web-based resource.

The SDK stored on the messaging server 118 effectively provides a bridge between external resources (e.g., third party or external applications 109 or applets and the messaging client 104). This provides the user with a seamless experience of communicating with other users on the messaging client 104, while also preserving the look and feel of the messaging client 104. To bridge communications between the external resource and the messaging client 104, in some examples, the SDK facilitates communications between the external resource server 110 and the messaging client 104. In some examples, the webviewjavascript bridge running on the client device 102 establishes two unidirectional communication channels between an external resource and the messaging client 104. Messages are sent asynchronously between the external resource and the messaging client 104 via these communication channels. Each SDK function call is sent as a message and a callback. Each SDK function is implemented by constructing a unique callback identifier and sending a message with the callback identifier.

By using the SDK, not all information from the messaging client 104 is shared with the external resource server 110. The SDK restricts which information is shared based on the needs of external resources. In some examples, each external resource server 110 provides an HTML5 file corresponding to a web-based external resource to the messaging server 118. Messaging server 118 may add a visual representation of a web-based external resource (e.g., box art or other graphic) in messaging client 104. Once the user selects the visual representation through the GUI of the messaging client 104 or instructs the messaging client 104 to access the features of the web-based external resource, the messaging client 104 obtains the HTML5 file and instantiates the resource needed to access the features of the web-based external resource.

The messaging client 104 presents a graphical user interface (e.g., a login page or banner screen) for the external resource. During, before, or after presentation of the login page or banner screen, the messaging client 104 determines whether the initiated external resource has previously been authorized to access user data of the messaging client 104. In response to determining that the initiated external resource has been previously authorized to access user data of the messaging client 104, the messaging client 104 presents another graphical user interface of the external resource that includes functions and features of the external resource. In response to determining that the initiated external resource was not previously authorized to access the user data of the messaging client 104, after a threshold period of time (e.g., 3 seconds) of displaying the login page or banner screen of the external resource, the messaging client 104 slides up the menu for authorizing the external resource to access the user data (e.g., animates the menu as appearing from the bottom of the screen to the middle of the screen or other portion). The menu identifies the type of user data that the external resource is authorized to use. In response to receiving the user selection of the accepted option, messaging client 104 adds the external resource to a list of authorized external resources and allows the external resource to access the user data from messaging client 104. In some examples, the external resource is authorized by the messaging client 104 to access the user data according to the OAuth 2 framework.

Messaging client 104 controls the type of user data shared with the external resource based on the type of external resource that is authorized. For example, an external resource that includes a full-scale external application (e.g., a third party or external application 109) is provided with access to a first type of user data (e.g., a two-dimensional-only avatar of a user with or without different avatar characteristics). As another example, an external resource that includes a small-scale version of an external application (e.g., a web-based version of a third-party application) is provided with access to a second type of user data (e.g., payment information, a two-dimensional avatar of the user, a three-dimensional avatar of the user, and avatars having various avatar characteristics). Avatar characteristics include different ways of customizing the avatar's look and feel, such as different poses, facial features, clothing, and the like.

Data architecture

Fig. 3 is a schematic diagram illustrating a data structure 300 that may be stored in database 126 of messaging server system 108, according to some examples. While the contents of the database 126 are shown as including a plurality of tables, it should be understood that the data may be stored in other types of data structures (e.g., as an object-oriented database).

The database 126 includes message data stored in a message table 302. For any particular one of the messages, the message data includes at least message sender data, message recipient (or recipient) data, and a payload. Additional details regarding information that may be included in a message and included in message data stored in message table 302 are described below with reference to fig. 4.

The entity table 306 stores entity data and is linked (e.g., referenced) to the entity graph 308 and profile data 316. The entities for which records are maintained within entity table 306 may include individuals, corporate entities, organizations, objects, places, events, and so forth. Regardless of the entity type, any entity about which messaging server system 108 stores data may be an identified entity. Each entity is provided with a unique identifier as well as an entity type identifier (not shown).

The entity graph 308 stores information about relationships and associations between entities. Such relationships may be merely social, professional (e.g., working at a common company or organization), interest-based, or activity-based, for example.

The profile data 316 stores various types of profile data about a particular entity. The profile data 316 can be selectively used and presented to other users of the messaging system 100 based on privacy settings specified by a particular entity. Where the entity is a person, the profile data 316 includes, for example, a user name, a phone number, an address, settings (e.g., notification and privacy settings), and an avatar representation (or set of such avatar representations) selected by the user. The particular user may then selectively include one or more of these avatar representations in the content of the message transmitted via messaging system 100 and on the map interface displayed by messaging client 104 to other users. The set of avatar representations may include a "situation avatar" that presents a graphical representation of a situation or activity that a user may select to communicate at a particular time.

Where the entity is a community, the profile data 316 for the community may similarly include one or more avatar representations associated with the community in addition to the community name, members, and various settings (e.g., notifications) for the relevant community.

The database 126 also stores enhancement data, such as overlays or filters, in the enhancement table 310. The enhancement data is associated with and applied to the video (the video data is stored in the video table 304) and the image (the image data is stored in the image table 312).

In one example, the filter is an overlay that is displayed as an overlay over the image or video during presentation to the receiving user. The filters may be of various types, including filters selected from among a set of filters presented to the sending user by the messaging client 104 when the sending user is composing a message. Other types of filters include geo-location filters (also referred to as geo-filters) that can be presented to a sending user based on geo-location. For example, a geographic location filter specific to a nearby or special location may be presented within the user interface by the messaging client 104 based on geographic location information determined by a Global Positioning System (GPS) unit of the client device 102.

Another type of filter is a data filter that can be selectively presented to a sending user by the messaging client 104 based on other input or information collected by the client device 102 during the message creation process. Examples of data filters include the current temperature at a particular location, the current speed at which the sending user is traveling, the battery life of the client device 102, or the current time.

Other augmented data that may be stored within the image table 312 includes augmented reality content items (e.g., corresponding to application shots or augmented reality experiences). The augmented reality content item may be real-time special effects and sounds that may be added to an image or video. Each augmented reality experience may be associated with one or more marker images. In some examples, when a tagged image is determined to match a query image received from client device 102, a corresponding augmented reality experience (e.g., augmented data) for the tagged image is retrieved from image table 312 and provided to client device 102.

As described above, the augmented data includes augmented reality content items, overlays, image transformations, AR images, and similar items that refer to modifications that can be applied to image data (e.g., video or images). This includes real-time modifications that modify an image as it is captured using a device sensor (e.g., one or more cameras) of the client device 102 and then displayed on a screen of the client device 102 if modified. This also includes modifications to the stored content (e.g., video clips in the gallery that can be modified). For example, in a client device 102 with access rights to multiple augmented reality content items, a user may use a single video clip with multiple augmented reality content items to see how different augmented reality content items will modify the stored clip. For example, multiple augmented reality content items applying different pseudo-random motion models may be applied to the same content by selecting different augmented reality content items for the same content. Similarly, real-time video capture may be used with the modifications shown to show how video images currently captured by the sensor of the client device 102 will modify the captured data. Such data may simply be displayed on the screen without being stored in memory, or content captured by the device sensor may be recorded and stored in memory with or without modification (or both). In some systems, the preview feature may show how different augmented reality content items will look simultaneously within different windows in the display. This may enable, for example, multiple windows with different pseudo-random animations to be viewed on the display at the same time.

Thus, data and various systems using augmented reality content items or other such transformation systems using the data to modify content may involve detection of objects (e.g., faces, hands, bodies, cats, dogs, surfaces, objects, etc.), tracking of such objects as they leave, enter, and move around the field of view in video frames, and modification or transformation of such objects as they are tracked. In various examples, different methods for implementing such transformations may be used. Some examples may involve generating a three-dimensional mesh model of one or more objects and implementing the transformation within the video using the transformations and animated textures of the model. In other examples, tracking of points on an object may be used to place an image or texture (which may be two-dimensional or three-dimensional) at the tracked location. In a still further example, neural network analysis of video frames may be used to place images, models, or textures in content (e.g., images or frames of a video). Thus, augmented reality content items refer to both images, models and textures used to create transformations in content, and to additional modeling and analysis information required to implement such transformations with object detection, tracking and placement.

Real-time video processing may be performed using any kind of video data (e.g., video streams, video files, etc.) stored in the memory of any kind of computerized system. For example, a user may load a video file and save it in a memory of the device, or may generate a video stream using a sensor of the device. Furthermore, any object, such as a human face and parts of a human body, an animal or a non-living being (e.g. a chair, a car or other objects) can be processed using the computer animated model.

In some examples, when a particular modification is selected with the content to be transformed, the element to be transformed is identified by the computing device, and then the element to be transformed is detected and tracked if it is present in a frame of the video. The elements of the object are modified according to the modification request, thereby transforming the frames of the video stream. The transformation of the frames of the video stream may be performed by different methods for different kinds of transformations. For example, for a frame transform that refers primarily to a variant of an element of an object, feature points are computed for each element of the object (e.g., using an Active Shape Model (ASM) or other known methods). Then, a feature point-based mesh is generated for each of the at least one element of the object. The mesh is used to track subsequent stages of elements of the object in the video stream. During the tracking process, the mentioned grid of each element is aligned with the position of each element. Additional points are then generated on the grid. A first set of first points is generated for each element based on the modification request, and a set of second points is generated for each element based on the set of first points and the modification request. The frames of the video stream may then be transformed by modifying elements of the object based on the set of first points and the set of second points and the mesh. In such an approach, the background of the modified object may also be changed or distorted by tracking and modifying the background.

In some examples, the transformation to change some regions of the object using the elements of the object may be performed by calculating feature points of each element of the object and generating a mesh based on the calculated feature points. Points are generated on the grid and then various regions based on the points are generated. The elements of the object are then tracked by aligning the region of each element with the position of each of the at least one element, and the properties of the regions may be modified based on the modification request, thereby transforming the frames of the video stream. The properties of the mentioned regions can be transformed in different ways depending on the specific modification request. Such modifications may involve: changing the color of the region; removing at least a partial region from a frame of a video stream; including one or more new objects in the modification request based region; and modifying or distorting elements of the region or object. In various examples, any combination of such modifications or other similar modifications may be used. For some models to be animated, some feature points may be selected as control points for determining the entire state space for options for model animation.

In some examples of computer animated models that use face detection to transform image data, faces are detected on the image using a particular face detection algorithm (e.g., viola-Jones). An Active Shape Model (ASM) algorithm is then applied to the facial regions of the image to detect facial feature reference points.

Other methods and algorithms suitable for face detection may be used. For example, in some examples, features are located using landmarks that represent distinguishable points that are present in most images under consideration. For example, for facial landmarks, the location of the left eye pupil may be used. Secondary landmarks may be used if the initial landmark is not identifiable (e.g., if the person has eye patches). Such a landmark identification process may be used for any such object. In some examples, a set of landmarks forms a shape. The shape may be represented as a vector using the coordinates of the points in the shape. One shape is aligned to another shape using a similarity transform (allowing translation, scaling and rotation) that minimizes the average euclidean distance between shape points. The average shape is the average of the aligned training shapes.

In some examples, the landmark search is started from an average shape that aligns with the position and size of the face determined by the global face detector. Such a search then repeats the steps of: the tentative shape is suggested by adjusting the positioning of the shape points by template matching of the image texture around each point, and then conforming the tentative shape to the global shape model until convergence occurs. In some systems, individual template matches are unreliable, and the shape model pools the results of weak template matches to form a stronger overall classifier. The entire search repeats at each level of the image pyramid, from coarse resolution to fine resolution.

The transformation system may capture an image or video stream on a client device (e.g., client device 102) and perform complex image manipulations locally on the client device 102 while maintaining a suitable user experience, computing time, and power consumption. Complex image manipulations may include size and shape changes, mood transitions (e.g., changing a face from frown to smile), state transitions (e.g., aging a subject, reducing apparent age, changing gender), style transitions, graphical element applications, and any other suitable image or video manipulation implemented by a convolutional neural network that has been configured to be efficiently executed on the client device 102.

In some examples, a computer animation model for transforming image data may be used by a system in which a user may capture an image or video stream of the user (e.g., self-portrait) using a client device 102 having a neural network operating as part of a messaging client 104 operating on the client device 102. A transformation system operating within messaging client 104 determines the presence of a face within an image or video stream and provides a modification icon associated with a computer animated model to transform the image data, or the computer animated model may be presented in association with an interface described herein. The modification icon includes a change that may be the basis for modifying the user's face within the image or video stream as part of the modification operation. Once the modification icon is selected, the transformation system initiates a process of converting the user's image to reflect the selected modification icon (e.g., generating a smiley face on the user). Once the image or video stream is captured and the specified modification is selected, the modified image or video stream may be presented in a graphical user interface displayed on the client device 102. The transformation system may implement a complex convolutional neural network on a portion of the image or video stream to generate and apply the selected modifications. That is, once the modification icon is selected, the user may capture an image or video stream and be presented with the results of the modification in real-time or near real-time. Further, the modification may be persistent while the video stream is being captured, and the selected modification icon remains toggled. A neural network of machine teachings can be used to implement such modifications.

Presenting the modified graphical user interface performed by the transformation system may offer additional interaction options for the user. Such options may be based on an interface (e.g., initiated from a content creator user interface) for initiating content capture and selection of a particular computer animation model. In various examples, the modification may be persistent after the initial selection of the modification icon. The user may turn the modification on or off by tapping or otherwise selecting the face modified by the transformation system and storing it for later viewing or browsing to other areas of the imaging application. In the case of multiple faces being modified by the transformation system, the user may turn the modification on or off globally by tapping or selecting a single face that is modified and displayed within the graphical user interface. In some examples, individual faces in a set of multiple faces may be individually modified, or such modifications may be individually switched by tapping or selecting individual faces or a series of individual faces displayed within a graphical user interface.

Story table 314 stores data about a collection of messages and associated image, video, or audio data that are compiled into a collection (e.g., a story or gallery). Creation of a particular collection may be initiated by a particular user (e.g., each user whose records are maintained in entity table 306). A user may create a "personal story" in the form of a collection of content that has been created and transmitted/broadcast by the user. To this end, the user interface of messaging client 104 may include user-selectable icons to enable the sending user to add specific content to his or her personal story.

Collections can also constitute "live stories" that are collections of content from multiple users that are created manually, automatically, or using a combination of manual and automatic techniques. For example, a "live story" may constitute a curated flow of user-submitted content from various locations and events. A user whose client device has a location service enabled and is at a common location event at a particular time may be presented with an option to contribute content to a particular live story, e.g., via a user interface of messaging client 104. A live story may be identified to a user by messaging client 104 based on his or her location. The end result is a "live story" that tells from a social perspective.

Another type of content collection is referred to as a "positioning story" that enables users whose client devices 102 are located within a particular geographic location (e.g., on a college or college campus) to contribute to a particular collection. In some examples, the contribution to the location story may require a second degree of authentication to verify that the end user belongs to a particular organization or other entity (e.g., is a student of a university campus).

As mentioned above, the video table 304 stores video data that, in one example, is associated with messages whose records are maintained within the message table 302. Similarly, the image table 312 stores image data associated with messages whose message data is stored in the entity table 306. Entity table 306 may associate various enhancements from enhancement table 310 with various images and videos stored in image table 312 and video table 304.

Data communication architecture

Fig. 4 is a schematic diagram illustrating the structure of a message 400, the message 400 generated by a messaging client 104 for transmission to another messaging client 104 or messaging server 118, according to some examples. The contents of a particular message 400 are used to populate a message table 302 stored in the database 126, the message table 302 being accessible by the messaging server 118. Similarly, the contents of the message 400 are stored in memory as "in-flight" or "in-flight" data for the client device 102 or the application server 114. Message 400 is shown to include the following example components:

the message identifier 402: a unique identifier that identifies the message 400.

Message text payload 404: text to be generated by a user via a user interface of the client device 102 and included in the message 400.

Message image payload 406: image data captured by a camera component of the client device 102 or retrieved from a memory component of the client device 102 and included in the message 400. Image data for a transmitted or received message 400 may be stored in the image table 312.

Message video payload 408: the video data captured by the camera component or retrieved from the memory component of the client device 102 and included in the message 400. Video data for a transmitted or received message 400 may be stored in the video table 304.

Message audio payload 410: audio data captured by a microphone or retrieved from a memory component of the client device 102 and included in the message 400.

Message enhancement data 412: enhancement data (e.g., a filter, sticker, or other annotation or enhancement) representing an enhancement to be applied to the message image payload 406, the message video payload 408, or the message audio payload 410 of the message 400. Enhancement data for a transmitted or received message 400 may be stored in the enhancement table 310.

Message duration parameter 414: a parameter value that indicates an amount of time in seconds for which the content of the message (e.g., message image payload 406, message video payload 408, message audio payload 410) is to be presented to or accessible by the user via the messaging client 104.

Message geolocation parameter 416: geolocation data (e.g., latitude and longitude coordinates) associated with the content payload of the message. A plurality of message geolocation parameter 416 values may be included in the payload, each of these parameter values being associated with a content item included in the content (e.g., a particular image in the message image payload 406, or a particular video in the message video payload 408).

Message story identifier 418: an identifier value identifying one or more content collections (e.g., "stories" identified in story table 314) with which a particular content item in message image payload 406 of message 400 is associated. For example, a plurality of images within message image payload 406 may each be associated with a plurality of sets of content using identifier values.

Message tag 420: each message 400 may be tagged with a plurality of tags, each of such plurality of tags indicating the subject matter of the content included in the message payload. For example, where a particular image included in message image payload 406 depicts an animal (e.g., a lion), a tag value may be included within message tag 420 that indicates the relevant animal. The tag value may be generated manually based on user input or may be generated automatically using, for example, image recognition.

Message sender identifier 422: an identifier (e.g., a messaging system identifier, an email address, or a device identifier) indicating the user of the client device 102 on which the message 400 was generated and from which the message 400 was sent.

Message recipient identifier 424: an identifier (e.g., a messaging system identifier, an email address, or a device identifier) indicating the user of the client device 102 to which the message 400 is addressed.

The content (e.g., values) of the components of message 400 may be pointers to locations in the table where content data values are stored. For example, the image value in the message image payload 406 may be a pointer to a location within the image table 312 (or an address of a location within the image table 312). Similarly, values within message video payload 408 may point to data stored within video table 304, values stored within message enhancement data 412 may point to data stored within enhancement table 310, values stored within message story identifier 418 may point to data stored within story table 314, and values stored within message sender identifier 422 and message recipient identifier 424 may point to user records stored within entity table 306.

Fig. 5 illustrates an overview of a geometry 500 of a camera device that captures two-dimensional images of objects in a three-dimensional scene, according to some examples. The illustrated example shows only the y-axis (vertical) and z-axis (depth) axes of a three-dimensional scene; the x-axis (extending horizontally into the page).

The camera 502 is located at a first position and captures an image of an object 504 (e.g., a human) in an image plane 506. The image plane is located at a focal distance f from the image pickup device 502 in the depth direction. The object has a height H in the three-dimensional scene and is positioned at a depth Z from the camera. In the image plane 506 of the camera 502, the object has a height h.

Object 504 and a position P = [ X, Y, Z ] located in a three-dimensional scene]Subject keypoints 508 (e.g., the pelvis or sternum of a human) of (a) are associated. A ray 510 (shown as a dashed line) connects the camera 502 to the keypoint 508 and a position p = [ x, y, f ] in the three-dimensional scene]Where it intersects the image plane. This corresponds to a pixel position I = (I) in a two-dimensional image captured by the imaging device _x ，i _y ). Principal point c = (c) of image pickup device 502 _x ，c _y ) The distance from the intersection is given by d.

The image capture device sampling density C can be used to convert the object size h in the image plane to the pixel size s in the two-dimensional image. Using this, the pixel size is given by s = hC. Using a similar triangle, H can be related to Z using the following formula:

rearranging the equation according to the pixel height s of the object, which gives:

thus, if H is known (e.g., by knowing a priori the dimensions of the frequent object, or from the output of the object model), then if f is also known, an estimate of Z can be obtained from s. However, given a test dataset for an image, typically f and/or C are unknown for the image.

This drawback can be overcome by training the model using "absolute depth" and converting the absolute depth to "world depth" Z at the extrapolated time when knowledge of C and f is available, where "absolute depth" is defined by:

the absolute depth depends only on the actual height H of the object and the pixel height s of the object. The pixel height may be determined directly from the image, while the actual height may be estimated based on a model of the object. For (adult) humans, height is typically limited to a relatively narrow range and can be estimated using, for example, the BLSM model. In some examples, a convolutional neural network (or a branch of any of the networks described herein) may be used to determine the pixel height s of the object.

In some implementations, the CNN (or a branch of any of the CNNs described herein) can be used to determine an intersection point I = (I) in a two-dimensional image captured by a camera device _x ，i _y ) The pixel coordinates of (a). The pixel coordinates may be converted to the keypoint 508 location using the following equation:

i _x -c _x = Cx and i _y -c _y ＝Cy

And is provided with

And

obtaining:

X＝(i _x -c _x )Z _abs

and

Y＝(i _y -c _y )Z _abs

then, Z is bonded _abs P is obtained from the formula:

P[(i _x -c _x )Z _abs ，(i _y -c _y )Z _abs ，CfZ _abs ]

note that only the depth Z depends on the focal length of the image pickup device.

Using current methods (e.g., BLSM networks for humans), a 3D representation of an object in a 2D image can be created using, for example, a trained model/neural network. BLSM networks are discussed in more detail in commonly owned U.S. patent application No. 16/949,781, filed 11/13/2020, which is incorporated herein by reference in its entirety. The representation is then positioned in the 3D scene using the determined position P of the object in the scene. For example, for a 3D representation in the form of a mesh in a mesh coordinate system (e.g., relative to a key point of the object) with a vertex { Bi }, it can be converted to a vertex in a world coordinate system using the following equation:

V _i ＝B _i +P

however, such methods do not take into account distortion/perspective shift of objects in the 2D image, especially for objects located near the edges of the 2D image, or if the 2D image is taken with a low focal length, resulting in a wrong positioning/orientation of the 3D representation in the 3D scene. This in turn may cause the 3D representation of the object to be erroneously projected back into the 2D image plane, which in turn may adversely affect the training of the model used to determine the 3D representation from the 2D image.

These problems are particularly prevalent when using convolutional neural networks, since the translational invariance of the network cannot account for position-dependent distortions in the image.

Fig. 6A illustrates an example of a method of reconstructing a three-dimensional object in a three-dimensional scene from a two-dimensional image, according to some examples. Fig. 6A shows an example of a method 600 of reconstructing a three-dimensional object in a three-dimensional scene from a two-dimensional image containing the object, which may account for such distortion/perspective shifts by applying a position-dependent rotation to the three-dimensional representation of the object as a post-processing step.

A 2D image 602 is received that includes an image of an object 604 (in this example, a human being). The neural network 606 is used to generate a 3D representation 608 of the object 604. The absolute depth 610 of the subject's 604 keypoints (e.g., sternum, pelvis, etc.) is also estimated from the 2D image 602 using the neural network 606, another neural network (not shown), or by some other model. The absolute depth is used to estimate the location 612 (i.e., P) of the keypoint in the 3D scene, for example, using the method described above with respect to fig. 5. A rotational transformation 614 (i.e., R) is determined from the determined position 612 and applied to the 3D representation 608 to create a rotated 3D representation 616. The rotated 3D representation is then positioned in the reconstructed 3D scene 618 using the locations 612 of the keypoints.

The neural network 606 takes the 2D image as input and generates a 3D representation of the object in the image from the image. One example of such a network 606 is the BLSM network described above. The neural network 606 may be a convolutional neural network. The neural network 606 may be a fully convolutional neural network. Other suitable types of machine learning models may be used instead of or in addition to the neural network 606.

The 3D representation 608 may be any type of 3D representation known in the art. For example, the 3D representation 608 may be in the form of a mesh comprising a plurality of vertices. Other possible 3D representations 608 include, but are not limited to, directed or undirected graphs.

Rotation transform 614 applies a position dependent rotation to the 3D representation of object 604. The rotation transform 614 is determined based on the keypoint location P (defined as described above with respect to fig. 5). For example, the rotation angle θ about the y-axis of the object (i.e., in the x-z plane) may be given by:

similar expressions apply mutatis mutandis to rotation around the x-axis of the object (i.e. in the y-z plane).

In some examples, the focal length of the camera that captured the image 602 may be left as a free variable that is determined at the inferred time by fitting a re-projection (e.g., the re-projection image 626 described below with respect to the training neural network 606) to the original image using a loss function (e.g., the loss function 628 described below with respect to the training neural network 606).

The 3D scene 618 generated by the method 600 may be used to train the neural network 606 to correctly recreate a 3D representation of the object from the 2D images. An example of a method 620 of training the neural network 606 is shown in FIG. 6B.

Fig. 6B illustrates an example of a method for training a neural network to correctly reconstruct a three-dimensional representation of an object from two-dimensional images, according to some examples. The method proceeds similarly to fig. 6A: a 2D training image 622 comprising images of objects 604 of a given type (in this example humans) is received from a training set of 2D images containing objects of that object type. The neural network 606 is used to generate a 3D representation 608 of the object 604 in a 2D training image 622. The absolute depths 610 of the key points (e.g., sternum, pelvis, etc.) of the object 604 are also estimated from the 2D training image 622 using the neural network 606, another neural network (not shown), or by some other model. The absolute depth is used to estimate the location 612 (i.e., P) of the keypoint in the 3D scene 618, e.g., using the method described above with respect to fig. 5. A rotational transformation 614 (i.e., R) is determined from the determined position 612 and applied to the 3D representation 608 to create a rotated 3D representation 616. The rotated 3D representation is then positioned in the reconstructed 3D scene 618 using the locations of the keypoints.

A 3D representation 624 of the model in reconstructed 3D scene 618 is projected into an image plane to generate a re-projection image 626. The re-projected image 626 is compared to the 2D training image 622, for example using a loss function 628, to determine updates to the parameters of the neural network 606. For example, the loss function 628 may compare the location of a point of the object in the re-projected image 626 to a corresponding point in the training image 602. The points used for comparison may be key points of the object 604, such as facial features and/or joints. The penalty function 628 may be an L1 or L2 penalty between these locations, but it will be understood that other examples are possible.

Parameter updates may be determined by applying an optimization procedure to the loss function 628. An example of such an optimization process is a random gradient descent, but it will be understood that other examples are possible. Parameter updates may be determined and applied after processing a batch of training images.

Fig. 7 shows an example of a method 700 for estimating a three-dimensional pose of an object (e.g., a human) in an image 702. A 3D pose is an example of a 3D representation of an object. Method 700 is an iterative method, one iteration being shown in FIG. 7. Method 700 may be performed by a computer.

The 3D pose is in the form of a directed graph comprising a plurality of nodes, where pairs of nodes are connected by edges. Each edge may have a direction associated with it, i.e., starting at one node and ending at another node. The map may be centered on key points of the body (e.g., sternum). In some examples, the directed graph may be star-shaped. In other examples, the graph may be a tree structure, where the mid-range edges are tightly connected by points on the human skeleton. The plurality of nodes may correspond to key points of the object, such as joints of the body.

The directed graph is iteratively constructed starting from the root node 704 of the graph. In each iteration, the new node 706 is added to the graph based on the depth of the previous node 710, the disparity 712 estimated near the location of the previous node 710 of the directed graph, and the estimated offset field 714 associated with the edge connecting the previous node 710 to the new node 706. The offset field 714 is estimated using one or more CNNs 716. One or more additional CNNs 718 (which may be a separate neural network or additional branches of one or more CNNs 716) are used to estimate the disparity near the location of the previous node 710. This is repeated until all nodes in the graph have been added.

Disparity is a measure of the rate of change of the corresponding projected coordinates of an object with world coordinates, i.e., for a position (X) in 3D _i ，Y _i ，Z _i ) A point of (x) whose 2D projection is _i ，y _i ) Parallax D _i From f/Z _i It is given.

The location of the root node 704 may be determined at the beginning of the method 700 by identifying the 2D location of the root node 704 in the image 702 and using one or more additional CNNs 718 to determine disparities in the region of the 2D image 702. The 2D location of the root node 706 in the image may be identified using, for example, a CNN representation based on a heat map. The root node 704 may correspond to a predetermined keypoint of the subject, such as the sternum of a human being. Alternatively, the root node 704 may be selected based on key points of objects visible in the 2D image, e.g., if the sternum is not visible, a joint of a human may be used as the root node 704.

One or more CNNs 718 used to estimate disparity have been trained to correctly compute disparity values in the vicinity of human keypoints using a large dataset of images labeled with 3-D keypoints, or using equivalent information obtained by fitting a parametric body model onto 2-D images, or a combination thereof.

One or more CNNs 716 used to estimate the offset field 714 may be divided into 2D branches and depth branches. The 2D branches provide a 2D flow field in the x-y plane, i.e., f ^e _2D ＝(f ^e _x ，f ^e _y ) While the depth branch provides an offset field in the depth direction, i.e. f ^e _z 。

For the 2D coordinates (i.e., coordinates in the x-y plane) of the new node 706, a 2D pose estimation method may be used. Two positions (x) of the given ith node _i ，y _i ) An edge e connecting the jth node to the ith node may be used _ij The associated 2D offset field, estimates the 2D position of the jth node by adding the 2D offset field to the 2D position of the ith node. 2-D offset field f ^e _2D Predicted by the 2D branches of one or more CNNs 716, where the CNNs 716 have been specifically trained to connect neighboring points in the object (person)'s motion tree graph using large datasets of images labeled with 2-D keypoints.

The disparity 712 estimated by one or more additional CNNs 716 and the depth offset estimated by one or more branches of the CNN 716 may be used, by shifting the depth by f ^e _z The ratio to disparity is added to the depth coordinate of the current node 706 to estimate the depth coordinate of the new node 706.

This modularity allows the CNN 716 to be more easily trained, for example, using a combination of 2-D keypoint data sets (2-D keypoint data sets are more abundant and easy to create using human annotations on images used publicly) and 3-D keypoint data sets (3-D keypoint data sets are more difficult to create and are typically limited to motion capture systems in indoor environments).

In some examples, a global depth correction may be applied to the determined location of the new node 706.

In some examples, the mesh nodes include keypoint nodes each associated with a keypoint of the object/human and leaf nodes each associated with a point on the surface of the human. Leaf nodes are each connected to one or more nearest keypoint nodes by directed edges. During the method, the three-dimensional coordinates of the keypoint nodes are first determined before the three-dimensional coordinates of the leaf nodes are determined, i.e., the "skeleton" of the model is first determined, followed by the surface.

Fig. 8 shows an example of another method 800 for estimating a three-dimensional pose of an object (e.g., a human) in an image 804. A 3D pose is an example of a 3D representation of an object. Method 800 is an iterative method. Method 800 may be performed by a computer.

In the method, a 3D representation of an object (e.g., a human) is defined according to a set of joints, each joint associated with a joint pose comprising a joint position and a joint rotation.

A set of known joint positions 802, i.e., P, corresponding to joints of an object (e.g., a human) in an image 804 is received ^k . Starting from the root joint 806, a model of the object is recursively fitted to the known joint positions 802. The model provides a set of constraints on allowable joint poses that are not present in the known joint positions 4802. For each joint, and starting from the root joint 806, the joint pose of the joint is determined based on a recurrence relation 808 that defines the joint pose of the joint according to the joint pose of the previous joint, the predefined (i.e., model-defined) offset between the joint and the previous joint, the bone rotation, and the bone scaling (related to the bone length). Bone scaling and/or bone rotation are variables used to fit the model to the known joint positions 802.

The set of known joint positions 802 may be, for example, ground-true (ground-truth) positions (e.g., obtained by a motion capture method). Alternatively, the known locations may be locations output by a "bottom-up" model (e.g., a CNN trained to identify joint locations in an image) or a disparity-based approach such as that described above with respect to fig. 7. The set of known joint positions 802 may also be referred to as a "world target" or "target position. Each known joint position may be denoted as w _i 。

In some examples, the known joint position 802 may be obtained by combining predictions from the subject site model. In these examples, as shown in fig. 9, one or more (e.g., a plurality of) 3D representations of the object site 810 are combined with the locations of the keypoints 812 of the object (also referred to as 3D representations of the keypoints) to generate the known joint locations 802.

A 3D representation of the object region is determined using the region-specific models 814, respectively. The part-specific model is a "bottom-up" model, such as a CNN trained to reproduce a 3D model/representation of a particular object part 810 (e.g., hand, foot, head, face, torso, etc.) from the image 804 containing the object. Each subject site may be associated with a different CNN or a different branch of a CNN. The 3D representation of the object region 810 may be in the form of a mesh of object regions. A 3D representation of the subject portion 810 may be defined in a coordinate system relative to key points of the subject portion (e.g., wrist joint for hand, sternum for torso, ankle joint for foot, facial features for head/face).

A keypoint location method 816 can be used to identify the location of the object's keypoints 812. The locations of the key points of the object include the 3D locations/coordinates (e.g., joint locations) of key features of the object. An example of such a method is a disparity-based keypoint method, such as the method described with respect to fig. 7. In some examples, CNN may be used to identify the location of keypoint 812.

One or more (e.g., multiple) 3D representations of the subject part 810 are combined with the location of the subject's keypoints 812 to generate known joint locations 802. Keypoints in the object portion 3D representation of object portion 810 may be matched to keypoints in the 3D representation of keypoint 812 and positioned in 3D space accordingly. Any of the methods described with respect to fig. 6A may be used to locate the 3D object region 810 in a manner that accounts for distortion in the 2D image 804, i.e., position dependent rotation may be applied. Alternatively, the 3D object sites 810 may simply be placed at the respective key points.

The resulting known joint positions 802 form a coarse 3D representation of the entire object, which positions the object part at the correct depth and may take into account distortion/perspective effects in the input 2D image 804. However, it may not properly join the different object regions 810 and may not comply with the object constraints. Thus, the parameterized 3D representation/model is recursively fitted to the known joint positions 802 to determine an overall 3D representation of the object that does obey the constraints but also has accurate joint positions. This may close the gap between model-based approaches for monocular 3D body reconstruction and CNN-based prediction of 3D body joints.

The parameterized 3D representation of the object comprises a plurality of joints, each joint and a joint pose P _i And (4) associating. The joint posture includes a joint position t _i (i.e., 3D position of joint in space) and joint rotation R _i (i.e., the 3D orientation of the joint in space, e.g., defined by a 3 x 3 rotation matrix). The posture canTo be represented by a 4 x 4 matrix:

starting from the root joint 806 (i.e., i = 0), the joint pose of the current joint j is determined from the joint pose of the previous joint i using a recurrence relation 808. First, if applied to the joint pose P of the previous joint _i Then a known (i.e., model-defined) offset O is applied _j . Known offsets encode fixed relationships between joints, e.g., constraints on the position and rotation of the joints relative to each other. The offset may comprise a position offset d _j And a rotational offset E _j . The offset may be represented by a 4 x 4 matrix:

the product of the joint pose of the previous joint and the offset matrix Mi converts the relative displacement/rotation of the joint into an absolute position.

Transforming the variable matrix U _i Applied to the product of the joint pose of the previous joint and the offset matrix to give the current node P _j The posture of the joint. The variable matrix includes bone scaling s _j And bone rotation matrix B _j I.e. it is a combination of rotation and scaling transformations. This can be symbolically represented as:

P _j ＝M _j U _j wherein, in the step (A),

thus, in these examples, the recursive relationship 808 of joint position and joint rotation is given by:

R _j ＝R _i E _j sjB _j

and

t _j ＝t _i +R _i d _j

the skilled person will appreciate that equivalent expressions may alternatively be used.

The bone rotation angle is used as a parameter to fit the parameterized 3D representation to the known joint positions 802. Bone scaling may also be used as an additional parameter for fitting, or may be fixed by a parameterized model.

In some implementations, at least one of the joints in the parameterized 3D representation is a joint with a single sub-joint k (i.e., with only one additional joint depending on its joint pose). The bone scaling of each such joint may be determined based on a predefined length between the joint and a sub-joint of the joint, for example by using known distances between the joint and its sub-nodes to ensure that the distances between the joint and its sub-nodes are correct. The known length between a joint and a sub-joint may be based on the joint w _j Is known position and joint child node w _k Is determined from the difference between the known positions. One or more bone rotation angles of such joints may be determined by ensuring that the position of the sub-joint matches the known position of the sub-joint.

In some implementations, at least one of the joints in the three-dimensional model is a multi-joint to which a plurality of sub-joints are attached. Thus, the bone rotation matrix and bone scaling are overdetermined for such joints. Thus, the bone rotation and/or bone scaling for each multi-joint may be determined based on minimizing the difference between the estimated sub-node positions determined using the known (i.e., target) joint positions for the multi-joint. The method is repeated for each joint in the subject until all joint poses are determined.

Fig. 10 is a flow diagram illustrating example operations of the messaging client 104 in performing the process 1000, according to an example. Process 1000 may be implemented with computer readable instructions for execution by one or more processors such that the operations of process 1000 may be performed in part or in whole by the functional components of messaging server system 108; accordingly, process 1000 is described below by way of example with reference to functional components of messaging server system 108. However, in other examples, at least some of the operations of process 1000 may be deployed on various other hardware configurations. The operations in process 1000 may be performed in any order, in parallel, or may be skipped and omitted entirely.

At operation 1001, the image processing server 122 determines absolute depths of key points of an object in an image from a two-dimensional image.

At operation 1002, the image processing server 122 determines a three-dimensional location of the keypoint in the three-dimensional scene using the absolute depth of the keypoint.

At operation 1003, the image processing server 122 generates a three-dimensional representation of the object using the neural network, the three-dimensional representation including a plurality of mesh nodes defined in a coordinate system relative to the keypoints.

At operation 1004, the image processing server 122 positions a three-dimensional representation of the object in the three-dimensional scene based at least in part on the locations of the keypoints.

At operation 1005, the image processing server 122 applies the position-dependent rotation to the three-dimensional object.

Machine architecture

Fig. 11 is a diagrammatic representation of a machine 1100 in which instructions 1108 (e.g., software, a program, an application, an applet, an application (app), or other executable code) may be executed to cause the machine 1100 to perform any one or more of the methodologies discussed herein. For example, the instructions 1108 may cause the machine 1100 to perform any one or more of the methods described herein. The instructions 1108 transform the general purpose non-programmed machine 1100 into a specific machine 1100 programmed to perform the functions described and illustrated in the described manner. The machine 1100 may operate as a standalone device or may be coupled (e.g., networked) to other machines. In a networked deployment, the machine 1100 may operate in the capacity of a server machine or a client machine in server-client network environment, or as a peer machine in a peer-to-peer (or distributed) network environment. The machine 1100 may include, but is not limited to: a server computer, a client computer, a Personal Computer (PC), a tablet computer, a laptop computer, a netbook, a set-top box (STB), a Personal Digital Assistant (PDA), an entertainment media system, a cellular telephone, a smart phone, a mobile device, a wearable device (e.g., a smart watch), a smart home device (e.g., a smart appliance), other smart devices, a network device, a network router, a network switch, a network bridge, or any machine capable of sequentially or otherwise executing instructions 1108 that specify actions to be taken by machine 1100. Further, while only a single machine 1100 is illustrated, the term "machine" shall also be taken to include a collection of machines that individually or jointly execute the instructions 1108 to perform any one or more of the methodologies discussed herein. For example, machine 1100 may comprise client device 102 or any of a number of server devices that form part of messaging server system 108. In some examples, the machine 1100 may also include both a client system and a server system, where certain operations of a particular method or algorithm are performed on the server side, and where certain operations of a particular method or algorithm are performed on the client side.

The machine 1100 may include a processor 1102, a memory 1104, and input/output (I/O) components 1138 that may be configured to communicate with one another via a bus 1140. In an example, processor 1102 (e.g., a Central Processing Unit (CPU), a Reduced Instruction Set Computing (RISC) processor, a Complex Instruction Set Computing (CISC) processor, a Graphics Processing Unit (GPU), a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Radio Frequency Integrated Circuit (RFIC), another processor, or any suitable combination thereof) may include, for example, a processor 1106 that executes instructions 1108 and a processor 1110. The term "processor" is intended to include multi-core processors that may include two or more independent processors (sometimes referred to as "cores") that may execute instructions simultaneously. Although fig. 11 shows multiple processors 1102, the machine 1100 may include a single processor with a single core, a single processor with multiple cores (e.g., a multi-core processor), multiple processors with a single core, multiple processors with multiple cores, or any combination thereof.

The memory 1104 includes a main memory 1112, a static memory 1114, and a storage unit 1116, which are all accessible to the processor 1102 via a bus 1140. The main memory 1104, static memory 1114, and storage units 1116 store instructions 1108 that implement any one or more of the methodologies or functions described herein. The instructions 1108 may also reside, completely or partially, within the main memory 1112, within the static memory 1114, within the machine-readable medium 1118 within the storage unit 1116, within at least one of the processors 1102 (e.g., within a cache memory of a processor), or within any suitable combination thereof during execution thereof by the machine 1100.

The I/O components 1138 may include various components for receiving input, providing output, generating output, transmitting information, exchanging information, capturing measurements, and the like. The particular I/O components 1138 included in a particular machine will depend on the type of machine. For example, a portable machine such as a mobile phone may include a touch input device or other such input mechanism, while a headless (headset) server machine may not include such a touch input device. It should be understood that I/O components 1138 may include many other components not shown in fig. 11. In various examples, I/O components 1138 may include user output components 1124 and user input components 1126. The user output components 1124 may include visual components (e.g., a display such as a Plasma Display Panel (PDP), a Light Emitting Diode (LED) display, a Liquid Crystal Display (LCD), a projector, or a Cathode Ray Tube (CRT)), acoustic components (e.g., speakers), tactile components (e.g., vibration motors, resistance mechanisms), other signal generators, and so forth. User input components 1126 may include alphanumeric input components (e.g., a keyboard, a touch screen configured to receive alphanumeric input, an optical keyboard, or other alphanumeric input components), point-based input components (e.g., a mouse, a touchpad, a trackball, a joystick, a motion sensor, or another pointing instrument), tactile input components (e.g., physical buttons, a touch screen providing location and/or force of touches or touch gestures, or other tactile input components), audio input components (e.g., a microphone), and so forth.

In another example, I/O components 1138 may include: biometric component 1128, motion component 1130, environmental component 1132, or location component 1134, among a wide variety of other components. For example, biometric component 1128 includes components for detecting expressions (e.g., hand expressions, facial expressions, vocal expressions, body gestures, or eye tracking), measuring bio-signals (e.g., blood pressure, heart rate, body temperature, perspiration, or brain waves), identifying a person (e.g., voice identification, retinal identification, facial identification, fingerprint identification, or electroencephalogram-based identification), and so forth. The motion component 1130 includes an acceleration sensor component (e.g., an accelerometer), a gravity sensor component, a rotation sensor component (e.g., a gyroscope).

The environment components 1132 include, for example: one or more camera devices (with still image/photo and video capabilities), lighting sensor components (e.g., photometer), temperature sensor components (e.g., one or more thermometers that detect ambient temperature), humidity sensor components, pressure sensor components (e.g., barometer), acoustic sensor components (e.g., one or more microphones that detect background noise), proximity sensor components (e.g., infrared sensors that detect nearby objects), gas sensors (e.g., gas detection sensors that detect concentrations of hazardous gases or measure pollutants in the atmosphere for safety), or other components that can provide indications, measurements, or signals corresponding to the surrounding physical environment.

As for the camera, the client apparatus 102 may have a camera system including, for example, a front camera on a front surface of the client apparatus 102 and a rear camera on a rear surface of the client apparatus 102. The front camera may, for example, be used to capture still images and video (e.g., "self-portrait") of a user of the client device 102, which may then be enhanced with the enhancement data (e.g., filters) described above. For example, a rear-facing camera may be used to capture still images and video in a more conventional camera mode, with these images similarly enhanced with enhancement data. In addition to front-facing cameras and rear-facing cameras, client device 102 may also include 360 ° cameras for capturing 360 ° photographs and videos.

Further, the camera system of the client device 102 may include dual rear cameras (e.g., a primary camera and a depth sensing camera), or even a triple, quadruple, or quintuple rear camera configuration on the front and back sides of the client device 102. For example, these multiple camera systems may include wide-angle cameras, ultra-wide-angle cameras, tele-cameras, macro-cameras, and depth sensors.

The location component 1134 includes a positioning sensor component (e.g., a GPS receiver component), an altitude sensor component (e.g., an altimeter or barometer that detects barometric pressure from which altitude may be derived), an orientation sensor component (e.g., a magnetometer), and the like.

Various techniques may be used to effect communication. I/O components 1138 also include communications components 1136, communications components 1136 operable to couple machine 1100 to network 1120 or device 1122 via a respective coupling or connection. For example, communications component 1136 may include a network interface component or other suitable device that interfaces with network 1120. In a further example of the present invention, communications component 1136 may include a wired communications component, a wireless communications component, a cellular communications component, a Near Field Communications (NFC) component, a,

The components (e.g.,

low power consumption),

Components, and other communication components that provide communication via other modalities. Device 1122 can be another machine or any of a variety of peripheral devices (e.g., a peripheral device coupled via USB).

Further, communications component 1136 can detect the identifier or include a component operable to detect the identifier. For example, communications component 1136 can include a Radio Frequency Identification (RFID) tag reader component, an NFC smart tag detection componentOptical reader components (e.g., optical sensors for detecting one-dimensional barcodes such as Universal Product Code (UPC) barcodes, multi-dimensional barcodes such as Quick Response (QR) codes, aztec codes, data matrices (Data Matrix), data symbols (Dataglyph), maximum codes (MaxiCode), PDF417, ultra codes (Ultra Code), UCC RSS-2D barcodes, and other optical codes) or acoustic detection components (e.g., microphones for identifying tagged audio signals). In addition, various information can be obtained via communications component 1136, such as location via Internet Protocol (IP) geolocation, via

Location via signal triangulation, location via detection of NFC beacon signals, etc., which may indicate a particular location.

Various memories (e.g., main memory 1112, static memory 1114, and memory of processor 1102) and storage units 1116 may store one or more sets of instructions and data structures (e.g., software) implemented or used by any one or more of the methodologies or functions described herein. In the case of execution by processor 1102, the instructions (e.g., instructions 1108) cause the various operations to implement the disclosed examples.

The instructions 1108 may be transmitted or received over the network 1120 via a network interface device (e.g., a network interface component included in the communications component 1136), using a transmission medium and using any one of several well-known transmission protocols (e.g., the hypertext transfer protocol (HTTP)). Similarly, the instructions 1108 may be transmitted or received via a coupling with the device 1122 (e.g., a peer-to-peer coupling) using a transmission medium.

Software architecture

Fig. 12 is a block diagram 1200 illustrating a software architecture 1204 that may be installed on any one or more of the devices described herein. The software architecture 1204 is supported by hardware, such as the machine 1202 including a processor 1220, memory 1226, and I/O components 1238. In this example, the software architecture 1204 may be conceptualized as a stack of layers, where each layer provides specific functionality. Software architecture 1204 includes layers such as operating system 1212, libraries 1210, framework 1208, and applications 1206. In operation, application 1206 invokes API call 1250 via the software stack and receives message 1252 in response to API call 1250.

The operating system 1212 manages hardware resources and provides common services. The operating system 1212 includes, for example, a kernel 1214, services 1216, and drivers 1222. The kernel 1214 serves as an abstraction layer between the hardware layer and other software layers. For example, the kernel 1214 provides memory management, processor management (e.g., scheduling), component management, networking, and security settings, among other functions. Services 1216 may provide other common services for other software layers. The driver 1222 is responsible for controlling or interfacing with the underlying hardware. For example, the driver 1222 may include a display driver, a camera driver,

Or

Low power drivers, flash drivers, serial communication drivers (e.g., USB drivers),

Drivers, audio drivers, power management drivers, and the like.

The library 1210 provides a common low-level infrastructure used by the applications 1206. The library 1210 may include a system library 1218 (e.g., a C-standard library), the system library 1218 providing functions such as memory allocation functions, string manipulation functions, mathematical functions, and the like. In addition, the libraries 1210 may include API libraries 1224, such as media libraries (e.g., libraries for supporting the presentation and manipulation of various media formats, such as moving picture experts group 4 (MPEG 4), advanced video coding (h.264 or AVC), moving picture experts group layer 3 (MP 3), advanced Audio Coding (AAC), adaptive multi-rate (AMR) audio codec, joint photographic experts group (JPEG or JPG), or Portable Network Graphics (PNG)), graphics libraries (e.g., openGL framework for presentation in two-dimensional (2D) and three-dimensional (3D) in graphics content on a display), database libraries (e.g., SQLite providing various relational database functions), web libraries (e.g., webKit providing web browsing functions), and so forth. The library 1210 may also include various other libraries 1228 to provide many other APIs to the application 1206.

The framework 1208 provides a common high-level infrastructure used by the applications 1206. For example, the framework 1208 provides various Graphical User Interface (GUI) functions, advanced resource management, and advanced location services. The framework 1208 may provide a wide variety of other APIs that may be used by the application 1206, some of which may be specific to a particular operating system or platform.

In an example, the applications 1206 can include a home application 1236, a contacts application 1230, a browser application 1232, a book reader application 1234, a location application 1242, a media application 1244, a messaging application 1246, a gaming application 1248, and a wide variety of other applications such as an external application 1240. The application 1206 is a program that executes functions defined in the program. One or more of the variously structured applications 1206 may be created in a variety of programming languages, such as an object-oriented programming language (e.g., objective-C, java, or C + +) or a procedural programming language (e.g., C language or assembly language). In a particular example, external application 1240 (e.g., using ANDROID by an entity other than the vendor of a particular platform) ^TM Or IOS ^TM Applications developed by Software Development Kit (SDK) may be in a software development environment such as IOS ^TM 、ANDROID ^TM 、

Mobile software running on Phone's mobile operating system or another mobile operating system. In this example, external applications 1240 may activate API calls 1250 provided by operating system 1212 to facilitate the functions described herein.

Glossary

"carrier wave signal" refers to any intangible medium that is capable of storing, encoding or carrying instructions for execution by the machine and includes digital or analog communication signals or other intangible medium to facilitate communication of such instructions. The instructions may be transmitted or received over a network using a transmission medium via the network interface device.

"client device" refers to any machine that interfaces with a communication network to obtain resources from one or more server systems or other client devices. The client device may be, but is not limited to, a mobile phone, desktop computer, laptop computer, portable Digital Assistant (PDA), smartphone, tablet computer, ultrabook, netbook, laptop computer, multiprocessor system, microprocessor-based or programmable consumer electronics, game console, set-top box, or any other communication device that a user may use to access a network.

A network, another type of network, or a combination of two or more such networks. For example, the network or a portion of the network may include a wireless network or a cellular network, and the coupling may be a Code Division Multiple Access (CDMA) connection, a global system for mobile communications (GSM) connection, or other type of cellular or wireless coupling. In this example, the coupling may implement any of various types of data transmission techniques, such as single carrier radio transmission technology (1 xRTT), evolution-data optimized (EVDO) techniques, general Packet Radio Service (GPRS) techniques, enhanced data rates for GSM evolution (EDGE) techniques, third generation partnership project (3 GPP) including 3G, fourth generation wireless (4G) networks, universal Mobile Telecommunications System (UMTS), high Speed Packet Access (HSPA), worldwide Interoperability for Microwave Access (WiMAX), long Term Evolution (LTE) standards, other data transmission techniques defined by various standards-setting organizations, other long-range protocols, or other data transmission techniques.

"component" refers to a device, physical entity, or logic having the following boundaries: the boundaries are defined by function or subroutine calls, branch points, APIs, or other techniques provided for partitioning or modularizing particular processing or control functions. A component may be combined with other components via its interfaces to perform a machine process. A component may be a packaged functional hardware unit designed for use with other components and may be part of a program that typically performs specific ones of the associated functions.

The components may constitute software components (e.g., code embodied on a machine-readable medium) or hardware components. A "hardware component" is a tangible unit that is capable of performing certain operations and may be configured or arranged in some physical manner. In various examples, one or more computer systems (e.g., a standalone computer system, a client computer system, or a server computer system) or one or more hardware components of a computer system (e.g., a processor or a group of processors) may be configured by software (e.g., an application or application portion) as a hardware component that operates to perform certain operations as described herein.

The hardware components may also be implemented mechanically, electronically, or in any suitable combination thereof. For example, a hardware component may comprise dedicated circuitry or logic that is permanently configured to perform certain operations. The hardware component may be a special purpose processor, such as a Field Programmable Gate Array (FPGA) or an Application Specific Integrated Circuit (ASIC). The hardware components may also include programmable logic or circuitry that is temporarily configured by software to perform certain operations. For example, the hardware components may include software executed by a general purpose processor or other programmable processor. Once configured by such software, the hardware component becomes a specific machine (or specific component of a machine) uniquely customized to perform the configured function, and is no longer a general purpose processor. It will be appreciated that the decision to implement a hardware component mechanically, in a dedicated and permanently configured circuit, or in a temporarily configured circuit (e.g., configured by software) may be driven by cost and time considerations. Accordingly, the phrase "hardware component" (or "hardware-implemented component") should be understood to encompass a tangible entity, that is, an entity that is physically constructed, permanently configured (e.g., hardwired), or temporarily configured (e.g., programmed) to operate in a certain manner or to perform certain operations described herein.

Considering the example of hardware components being temporarily configured (e.g., programmed), each of the hardware components need not be configured or instantiated at any one time. For example, where the hardware components include a general-purpose processor that is configured by software to be a special-purpose processor, the general-purpose processor may be configured at different times to be different special-purpose processors (e.g., including different hardware components), respectively. Thus, software configures one or more particular processors to constitute, for example, particular hardware components at one time and different hardware components at different times.

A hardware component may provide information to and receive information from other hardware components. Accordingly, the described hardware components may be considered to be communicatively coupled. Where multiple hardware components are present at the same time, communication may be achieved by signal transmission between or among two or more hardware components (e.g., through appropriate circuitry and buses). In examples in which multiple hardware components are configured or instantiated at different times, communication between such hardware components may be accomplished, for example, by storing information in a memory structure accessible to the multiple hardware components and retrieving the information in the memory structure. For example, one hardware component may perform an operation and store the output of the operation in a memory device communicatively coupled thereto. Other hardware components may then access the memory device at a later time to retrieve the stored output and process it. The hardware component may also initiate communication with an input device or an output device and may operate on the resource (e.g., collection of information).

Various operations of the example methods described herein may be performed, at least in part, by one or more processors that are temporarily configured (e.g., via software) or permanently configured to perform the relevant operations. Whether temporarily configured or permanently configured, such a processor may constitute processor-implemented components that operate to perform one or more operations or functions described herein. As used herein, "processor-implemented component" refers to a hardware component that is implemented using one or more processors. Similarly, the methods described herein may be implemented at least in part by processors, where a particular processor or processors are examples of hardware. For example, at least some of the operations of a method may be performed by one or more processors 1102 or processor-implemented components. Further, the one or more processors may also operate to support execution of related operations in a "cloud computing" environment or to operate as a "software as a service" (SaaS). For example, at least some of the operations may be performed by a set of computers (as an example of machines including processors), where the operations are accessible via a network (e.g., the internet) and via one or more appropriate interfaces (e.g., APIs). The execution of certain operations may be distributed among the processors, and may reside not only within a single machine, but also deployed over multiple machines. In some examples, the processors or processor-implemented components may be located in a single geographic location (e.g., within a home environment, an office environment, or a server farm). In other examples, the processor or processor-implemented component may be distributed across multiple geographic locations.

"computer-readable storage medium" refers to both machine storage media and transmission media. The term therefore includes both storage devices/media and carrier wave/modulated data signals. The terms "machine-readable medium," "computer-readable medium," and "device-readable medium" mean the same thing, and may be used interchangeably in this disclosure.

An "ephemeral message" refers to a message that is accessible for a time-limited duration. The ephemeral message may be text, an image, a video, etc. The access time for an ephemeral message may be set by the message sender. Alternatively, the access time may be a default setting or a setting specified by the recipient. Regardless of the setup technique, the message is temporary.

"machine-storage medium" refers to a single or multiple storage devices and media (e.g., a centralized or distributed database, and associated caches and servers) that store the executable instructions, routines, and data. Thus, the term should be taken to include, but not be limited to, solid-state memories, and optical and magnetic media, including memories internal or external to the processor. Specific examples of machine storage media, computer storage media, and device storage media include: non-volatile memory, including, for example, semiconductor memory devices, such as erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), FPGA, and flash memory devices; magnetic disks such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks. The terms "machine storage medium," "device storage medium," "computer storage medium" are intended to be the same thing, and may be used interchangeably in this disclosure. The terms "machine storage medium," computer storage medium, "and" device storage medium "expressly exclude a carrier wave, a modulated data signal, and other such media, at least some of which are encompassed by the term" signal medium.

A "non-transitory computer-readable storage medium" refers to a tangible medium capable of storing, encoding, or carrying instructions for execution by a machine.

"signal medium" refers to any intangible medium that is capable of storing, encoding or carrying instructions for execution by the machine, and includes digital or analog communications signals or other intangible medium to facilitate communication of software or data. The term "signal medium" shall be taken to include any form of modulated data signal, carrier wave, or the like. The term "modulated data signal" means a signal that has one or more of its characteristics set or changed in such a manner as to: the way in which information is encoded in the signal. The terms "transmission medium" and "signal medium" mean the same thing and may be used interchangeably in this disclosure.

Variations and modifications may be made to the disclosed examples without departing from the scope of the disclosure. These and other changes or modifications are intended to be included within the scope of the present disclosure as expressed in the following claims.

Claims

1. A computer-implemented method of reconstructing a three-dimensional model of an object in a three-dimensional scene from two-dimensional images, the two-dimensional images comprising images of the object, the method comprising:

determining, from the two-dimensional image, depths of keypoints of the object in the image;

determining a three-dimensional location of the keypoint in the three-dimensional scene using the depth of the keypoint;

generating a three-dimensional representation of the object using a neural network, the three-dimensional representation comprising a plurality of mesh nodes defined in a coordinate system relative to the keypoints; and

locating a three-dimensional representation of the object in the three-dimensional scene based at least in part on the locations of the keypoints,

wherein positioning the three-dimensional representation of the object comprises applying a position dependent rotation to the three-dimensional object.

2. The method of claim 1, wherein determining the location of the keypoint in the three-dimensional scene comprises:

determining an intersection point of a ray connecting a camera capturing the two-dimensional image to a keypoint of the object and an image plane of the two-dimensional image using the same neural network or another neural network; and

determining a location of the keypoint of the object in a coordinate system of the three-dimensional scene based on a world depth and the intersection.

3. The method of any of claims 1-2, wherein determining the location of the keypoint in the three-dimensional scene is further based on a focal length of a camera that captured the two-dimensional image.

4. The method according to any one of claims 1 to 3, wherein the position dependent rotation is a rotation according to the following angles: the angle depends on the ratio of two components of the three-dimensional position of the keypoint, wherein one of the two components is a depth component.

5. The method of any of claims 1 to 4, further comprising:

projecting the positioned three-dimensional representation into a two-dimensional image plane to generate a projection image of the object;

comparing the position of a point in the projected image with the position of a corresponding point in the two-dimensional image; and

updating a parameter of the neural network based on the comparison.

6. The method of claim 5, wherein projecting the located three-dimensional representation into a two-dimensional image plane is based at least in part on depths of mesh nodes defined by the three-dimensional representation and absolute depths of the keypoints.

7. A computer-implemented method of reconstructing a three-dimensional model of an object in a three-dimensional scene from a two-dimensional image, the two-dimensional image comprising an image of the object, the three-dimensional model comprising a directed graph comprising a plurality of nodes, each node being linked to one or more other nodes by one or more edges, the method comprising:

for each of a plurality of nodes in the directed graph:

estimating, using one or more convolutional neural networks and based on two-dimensional image coordinates of a previous node in the directed graph, a three-dimensional offset field associated with an edge connecting the node to the previous node;

estimating a disparity in the two-dimensional image in the vicinity of a point associated with the previous node using a further convolutional neural network; and

determining three-dimensional coordinates of the node using the three-dimensional coordinates of the previous node, the three-dimensional offset field, and the estimated disparity.

8. The method of claim 7, determining three-dimensional coordinates of the node comprising:

determining two-dimensional image coordinates of the node based on the two-dimensional image coordinates of the previous node and two components of the three-dimensional offset field;

determining a depth coordinate of the node based on the depth coordinate of the previous node, one of the components of the three-dimensional offset field, and the estimated disparity; and

converting two-dimensional image coordinates of the node to two-dimensional scene coordinates of the node using depth coordinates of the node or estimated disparity associated with the node.

9. The method of any of claims 7-8, wherein determining the depth coordinate of the node based on the depth coordinate of the previous node comprises:

determining an initial estimate of the depth coordinate based on depth coordinates of the previous node, components of the three-dimensional offset field, and the estimated disparity; and

refining the initial estimate of the depth coordinate by translating the nodes in a depth direction by a constant value, wherein the constant value is a weighted average of differences between the depth coordinate of each node and the disparity estimate of the depth of each node.

10. The method of any of claims 7 to 9, wherein the method starts from a root node and the method further comprises determining three-dimensional coordinates of the root node.

11. The method of claim 10, wherein determining three-dimensional coordinates of the root node comprises:

estimating two-dimensional coordinates of the root node from the two-dimensional image;

estimating a disparity near a point in the two-dimensional image associated with the root node using a further convolutional neural network; and

determining a depth coordinate of the root node based on the estimated disparity associated with the root node, the two-dimensional coordinates of the root node, and a focal length of a camera device that captured the two-dimensional image.

12. The method of any of claims 7 to 11, wherein the one or more convolutional neural networks comprise:

a first convolutional neural network configured to estimate a two-dimensional offset field in a plane associated with the two-dimensional image; and

a second neural network configured to estimate a one-dimensional depth offset field in a direction perpendicular to the plane.

13. The method of any of claims 7 to 12, wherein the plurality of nodes comprises:

a plurality of keypoint nodes, each keypoint node being associated with a keypoint of the object; and

a plurality of leaf nodes, each leaf node associated with a point on a surface of the model and each leaf node connected to one or more nearest keypoint nodes by directed edges,

wherein the three-dimensional coordinates of the keypoint node are determined before the three-dimensional coordinates of the leaf node are determined.

14. A computer-implemented method of reconstructing a three-dimensional model of an object from two-dimensional images, the three-dimensional model including a plurality of joints, each joint defined by a joint pose, the joint pose including a joint position and a joint rotation, the method comprising:

receiving a known joint position for each joint; and

for each joint, and starting from a root joint, determining a joint pose for the joint based on a recursive relationship defining the joint pose for the joint as a function of the joint pose for a previous joint, a predefined offset between the joint and the previous joint, bone rotation, and bone scaling,

wherein the bone rotation is determined based at least in part on a known joint position for each joint.

15. The method of claim 14, wherein the known joint position is a field true joint position.

16. The method of any one of claims 14 to 15, wherein the known joint positions are estimated joint positions obtained using a convolutional neural network.

17. The method of any one of claims 14 to 16, wherein the known joint positions are estimated by combining joint positions estimated by one or more object part models with estimated three-dimensional keypoint positions of the object.

18. The method of any of claims 14 to 17, wherein at least one of the joints in the three-dimensional model is a joint having a single sub-joint, and wherein bone scaling for each such joint is determined based on a predefined length between the joint and the sub-joint of the joint.

19. The method of any of claims 14 to 18, wherein the known lengths between the joint and the sub-joints of the joint are determined based on differences between known positions of the joint and known positions of the sub-nodes of the joint.

20. The method of any of claims 14 to 19, wherein at least one of the joints in the three-dimensional model is a multi-joint to which a plurality of sub-joints are attached, and wherein the bone rotation for each multi-joint is determined based on minimizing a difference between estimated sub-node positions determined using known joint positions for the multi-joint.